Self-Generated Whisker Movements Drive State-Dependent Sensory Input to Developing Barrel Cortex: Current Biology | Current Biology
James C. Dooley, Ryan M. Glanz, Greta Sokoloff, Mark S. Blumberg
Results
Previous studies have identified behavioral state as a factor that moderates the relationship between self-generated movements and neural activity. Specifically, in infant rats before postnatal day (P)11, whereas sensory feedback from limb movements during wake only weakly activates sensorimotor cortex, sensory feedback from limb twitches during active sleep triggers strong cortical responses [
Thalamic network oscillations synchronize ontogenetic columns in the newborn rat barrel cortex.
]. It is plausible, then, that by ignoring behavioral state, we underestimate the importance of sensory feedback from self-generated movements for driving early neural activity.
To address this issue directly, we recorded whisker movements during active sleep (AS) and wake (W) in P5 rats while recording extracellular activity in the C row of barrel cortex. All whiskers, except those in the C row, were trimmed. Whisker movements were recorded using high-speed video at 100 frames⋅s−1 and analyzed using DeepLabCut, a deep-learning method for markerless tracking of behavior (Figure 1A; [
DeepLabCut: markerless pose estimation of user-defined body parts with deep learning.
]). Local field potential (LFP) and spike trains of individual neurons were extracted from the raw neural data (Figures 1B–1D). LFP recordings were filtered for spindle bursts (8–40 Hz), the predominant oscillation observed in developing sensorimotor cortex [
(A) A video frame, shot from below, illustrating the method for tracking whisker movements using DeepLabCut. Colored dots are located at the tips of four C-row whiskers. Whisker displacement was measured along the animal’s anterior-posterior axis. At the bottom left of the image, the LED used for data synchronization is shown.
(B) Illustration of primary somatosensory cortex (S1) in a P5 rat. Whisker barrels are depicted in dark gray within the S1 map. Electrode shank recording sites for each animal are represented by red lines in the C row of the barrel field.
(C) Representative histological section of barrel cortex. The S1 barrel field was visualized using cytochrome oxidase. The electrode location for a single pup is shown within the yellow dashed box, which is enlarged in the inset to show the fluorescent electrode tracts.
(D) Representative 100 s record for an individual pup during active sleep (blue shading) and wake (red shading). From top to bottom: local field potential (LFP) recorded in barrel cortex. Unit activity recorded from barrel cortex; each row denotes a different single unit. Whisker displacement, in mm, of all tracked C-row whiskers (i.e., mean displacement). Nuchal EMG recording indicates periods of high muscle tone indicative of wake and periods of low muscle tone, punctuated by brief spikes (i.e., nuchal twitches), indicative of active sleep. Note that LFP and unit activity are highest during active sleep and that whisker movements coincide with increases in barrel cortex activity. See also Figure S1.
Animals slept for 32.7 ± 1.93 min during the 60-min recording sessions (Figure S1A). Whisker twitches and wake movements occurred at equal rates (Figure S1B; AS, 27.1 ± 1.45 min−1; W, 25.6 ± 2.28 min−1; t(7) = 0.43, p = 0.679, adj.
If whisker movements and barrel cortex activity are causally unrelated, we would expect them to sometimes occur together by chance. Therefore, the analyses below compare the relationship between whisker movements and barrel cortex activity with the expected relationship due to chance.
Whisker Movements Drive Neural Activity
We first measured the change in LFP power (2–100 Hz) before and after whisker twitches and wake movements. As shown in Figure 2A, peak LFP power increased significantly in the 500-ms period after whisker movements. Peak power was greater after twitches (7.8 ± 0.62 dB) than wake movements (5.8 ± 0.58 dB; t(7) = 4.19, p = 0.004, adj.
= 0.674). In addition, as shown in Figure 2B, the rate of spindle bursts was significantly higher during active sleep (7.9 ± 0.75 min−1) than during wake (3.3 ± 0.30 min−1; t(7) = 5.95, p = 0.001, adj.
= 0.811). The mean amplitude, duration, and peak frequency of spindle bursts produced during active sleep and wake did not differ significantly (data not shown).
Figure 2Whisker Movements Drive Spindle Bursts in Barrel Cortex
(A) Frequency spectrograms showing LFP activity, averaged across pups, in relation to the onset (at 0 s) of whisker twitches (top) and wake movements (bottom). The color axis represents LFP power, in dB above baseline.
(B) Mean (±SEM) rate of spindle bursts per minute during active sleep (blue circle) and wake (red circle). Gray lines show spindle burst rate for individual pups (n = 8). Asterisk denotes significant difference, p < 0.05.
(C) Illustration of the method used to determine the likelihood of a spindle burst given a whisker movement. Bottom: the black line represents the displacement of the whiskers as a function of time and the two gray tick marks below indicate the onset of whisker movements. Each movement onset initiates a 500-ms window, shown above. Top: LFP signal with two spindle bursts. The tick marks along the gray line below indicate the onset of a spindle burst. For this analysis, two key possibilities are shown: a whisker movement that is followed within 500 ms by a spindle burst (green rectangle) and a movement that is not followed within 500 ms by a spindle burst (gray rectangle).
(D) Mean (±SEM) likelihood of a spindle burst given a whisker twitch (blue line) or wake movement (red line) in relation to movement onset. The gray rectangle after movement onset shows a 500-ms window. Note that the peak increase in spindle burst likelihood for both twitches and wake movements falls within this 500-ms window.
(E) Mean (±SEM) likelihood of a spindle burst within 500 ms of a twitch (upper blue circle) or wake movement (upper red circle). The lower colored circles represent the mean (±SEM) likelihood of a spindle burst due to chance during active sleep (blue) and wake (red). Gray lines show data for individual pups (n = 8). The expected spindle burst likelihoods for individual animals are not shown. The black asterisk indicates a significant interaction (p Figure S2.
(F) Mean (±SEM) likelihood of a spindle burst given a twitch (blue line) or wake movement (red line) of a given amplitude. Whisker movements were sorted into one of ten bins of increasing amplitude (0–1 along the x axis), normalized to the maximum amplitude of whisker deflections observed in each pup. Bottom: the amount of whisker displacement, relative to the maximum displacement for each pup, depicted in green. The horizontal line denotes the threshold used to detect whisker movements (STAR Methods). The expected likelihood of a spindle burst for movements of any amplitude is depicted as a dashed blue (active sleep) or dashed red (wake) line. Black asterisk denotes a significant interaction (p Figure S3.
We next measured the likelihood of observing a spindle burst in barrel cortex in the 500-ms period after a whisker movement (Figure 2C). A 500-ms time window was chosen because it captures the period of increased spindle burst likelihood after a whisker movement (Figure 2D). After a whisker twitch, the observed likelihood of a spindle burst was 22.5% ± 2.5%, compared with the expected value of 6.7% ± 0.6% (Figure 2E, blue dots). In contrast, after a wake movement, the observed likelihood of a spindle burst was 11.9% ± 1.6%, compared with the expected value of 2.8% ± 0.3% (Figure 2E, red dots). Spindle bursts were significantly more likely to be triggered by twitches than wake movements, as evidenced by a significant interaction between behavioral state and the observed-expected likelihood difference (F(1, 7) = 8.62, p = 0.022, adj.
= 0.488). In summary, these results indicate that spindle bursts are twice as frequent during active sleep and 3–4 times more likely to occur after whisker movements than expected by chance.
Parallel analyses performed on single units in barrel cortex yielded similar results. Mean firing rates were higher during active sleep than wake (Figure S2A; t(36) = 6.31, p
= 0.512). After twitches, firing rates increased to a greater degree than after wake movements (Figure S2B; t(36) = 3.84, p
= 0.270). Additionally, the observed-expected likelihood of a firing rate increase was significantly greater after twitches than wake movements (Figure S2C; F(1, 36) = 10.39, p = 0.003, adj.
= 0.202). Finally, there was a strong relationship between spindle bursts and unit activity: when twitches or wake movements triggered spindle bursts, we observed substantially stronger single-unit responses than when movements did not trigger spindle bursts (Figures S2D and S2E).
Movement Amplitude Moderates Neural Activity
That a small minority of movements trigger a response in barrel cortex recently led researchers to conclude that barrel activity is mostly independent of self-generated movements (see [
Patchwork-type spontaneous activity in neonatal barrel cortex layer 4 transmitted via thalamocortical projections.
]). However, this conclusion is confounded by an important statistical issue: at these ages, whisker movements are relatively abundant (>25 min−1; Figure S1B) compared with neural activity (3–8 spindle bursts⋅min−1, Figure 2B; −1, Figure S2A). Clearly, many whisker movements do not trigger neural activity. However, not all whisker movements are the same; specifically, small whisker movements may be less likely than large ones to trigger barrel cortex activity.
Thus, we next analyzed whether movement amplitude moderates the likelihood of barrel cortex activity. To test this possibility, whisker movements were assigned to ten bins, from smallest to largest peak displacement. Small movements were much more frequent than large movements (Figure S3A), and the mean amplitude of twitches and wake movements was equal for each bin (Figure S3B).
The likelihood of a spindle burst occurring within the 500-ms period after a whisker twitch or wake movement was determined for each bin. Spindle burst likelihood in barrel cortex increased with movement amplitude (Figure 2F). The likelihood of a spindle burst occurring after the smallest-amplitude whisker twitches was 3.7% ± 0.4% and increased significantly with amplitude to 48.1% ± 8.3% (Figure 2F, blue line). For wake movements, the likelihood of a spindle burst also increased with amplitude, from 1.4% ± 0.2% to 20.6% ± 2.9% (Figure 2F, red line). This amplitude-dependent increase was greater for twitches than for wake movements, as evidenced by a significant interaction between behavioral state and movement amplitude (F(9, 63) = 2.34, p = 0.024, adj.
= 0.144). In contrast with movement amplitude, movement duration was not associated with increased spindle burst likelihood (data not shown).
We observed a similar relationship between movement amplitude and unit activity in barrel cortex (Figure S3C). The likelihood of unit activity increased with movement amplitude. This increase was stronger for twitches than wake movements, as evidenced by a significant interaction between behavioral state and movement amplitude (F(4.23, 152.39) = 4.67, p
= 0.090).
These results show that although large-amplitude movements represent a small proportion of all movements, they more reliably trigger barrel activity. However, even the largest amplitude movements trigger a spindle burst only 48% of the time during active sleep, and 21% of the time during wake. The probabilistic nature of sensory responses to movement is puzzling as it suggests that cortical responses are modulated both within and between behavioral states. Additional work is needed to characterize the modulatory mechanisms involved and determine their consequences for neural plasticity.
Neural Activity Is Reliably Preceded by Whisker Movements
As noted above, the likelihood of a whisker movement producing a spindle burst is necessarily low as a result of the high rate of movement and the low rate of neural activity. To more fully assess the relationship between movement and spindle bursts, we inverted the analysis and determined the likelihood that a given spindle burst was preceded by a movement (Figure 3A). By analogy, to understand the relationship between smoking (a relatively high-frequency event) and lung cancer (a relatively low-frequency event), one would assess both the likelihood that smokers will develop lung cancer and the likelihood that people with lung cancer smoked.
Figure 3Spindle Bursts Are Reliably Preceded by Whisker Movements
(A) Illustration of the method used to determine the likelihood of a whisker movement given a spindle burst. Illustration is the reverse of that in Figure 2C, with the triggering event being the onset of a spindle burst (bottom) rather than a whisker movement. For this analysis, two key possibilities are shown: a spindle burst that is preceded within 500 ms by a whisker movement (green rectangle) and a spindle burst that is not preceded within 500 ms by a whisker movement (gray rectangle).
(B) Mean (±SEM) likelihood of a whisker twitch (blue) or wake movement (red) in relation to the onset of a spindle burst. The gray rectangle before spindle burst onset shows a 500-ms window.
(C) Mean (±SEM) likelihood of a whisker twitch (upper blue circle) or wake movement (upper red circle) within the 500-ms window before a spindle burst. The lower colored circles represent the mean (±SEM) likelihood of a twitch (blue) or wake movement (red) due to chance. Gray lines show data for individual pups (n = 8). The expected movement likelihood for individual pups is not shown. Black asterisk indicates a significant interaction (p Figure S4.
(D) Mean rate of spindle bursts per minute during active sleep (blue bar) and wake (red bar) directly attributable to whisker movements (i.e., movement-related activity). Gray bars represent the rate of spindle bursts not attributable to whisker movement (i.e., residual activity).
As with the previous analyses, we observed the highest likelihood of a whisker movement within a 500-ms window before spindle burst initiation during both active sleep and wake (Figure 3B). The likelihood of a whisker movement preceding a spindle burst was significantly greater than the expected value due to chance (Figure 3C; F(1, 7) = 163.45, p
= 0.953). Whisker twitches preceded 55.0% ± 3.1% of all spindle bursts during active sleep, compared with the expected value of 25.0% ± 1.3%. Wake movements preceded 56.4% ± 2.9% of all spindle bursts during wake, compared with the expected value of 23.2% ± 2.0%. There was no effect of behavioral state: spindle bursts during sleep or wake were equally likely to be preceded by a twitch or wake movement, respectively. However, when expressed in raw numbers (Figure 3D), more than twice as many spindle bursts were preceded by twitches than wake movements.
We again observed similar results for unit activity (Figure S4). Proportionately, the percentage of unit activity preceded by twitches (53.8% ± 2.6%) was lower than the percentage of activity preceded by wake movements (60.4% ± 3.4%), indicated by a significant interaction between behavioral state and the observed-expected amount (F(1, 7) = 12.12, p = 0.010, adj.
= 0.582). Nevertheless, significantly more unit activity than expected was preceded by both twitches (t(7) = 13.70, p
= 0.964) and wake movements (t(7) = 13.17, p
= 0.956).
Neural Activity Increases during Periods of Movement
The 45% of residual activity identified in Figure 3D could arise if sensory feedback from movements in other parts of the body were able to drive activity in barrel cortex. Alternatively, the residual activity could arise spontaneously within the brain, independent of movement. To address these two possibilities, we used a time-based behavioral analysis that includes movements of the ipsilateral whiskers, hindlimbs, and tail. Specifically, we classified each time point into one of three movement categories based on whether (1) the contralateral whiskers were moving, (2) the contralateral whiskers were not moving but other body parts were moving, and (3) the pup was not detectably moving. Each period of movement began with the onset of the first movement and ended 500 ms after the last movement. As shown in Figure 4A, pups spent roughly half their time moving or not moving, and during periods when pups were moving, the contralateral whiskers were involved the majority of the time.
Figure 4Neural Activity Increases during Periods of Movement
(A) Mean percent of time during active sleep and wake when the contralateral whiskers were moving (dark blue/red), when other parts of the body were moving but the contralateral whiskers were not (light blue/red), and when no discernible movements were detected (white).
(B) Mean (±SEM) rate of spindle bursts per minute during active sleep (left, blue) and wake (right, red). Light blue/red lines show data for individual pups (n = 8). Dashed horizontal lines indicate the mean rate of spindle bursts across the entirety of active sleep and wake periods. Adjacent to each mean value, spindle burst rate is expressed as a percent change in relation to the mean rate within that state. Asterisks indicate significant differences (p < 0.05) between movement categories.
As shown in Figure 4B, the rate of spindle burst production was influenced both by movement category (F(2, 14) = 6.33, p
= 0.818) and behavioral state (F(1, 7) = 6.33, p = 0.040, adj.
= 0.400). For both active sleep and wake, periods of contralateral whisker movements were associated with the highest rates of spindle burst production (AS, 13.8 ± 1.48 min−1; W, 7.5 ± 0.98 min−1). Periods of other movements corresponded to an intermediate rate of spindle burst production (AS, 4.8 ± 0.51 min−1; W, 3.5 ± 0.82 min−1). Periods of no detectable movement were associated with the lowest rate of spindle burst production (AS, 2.5 ± 0.24; W, 0.8 ± 0.11 min−1). Because periods of other movements elevated the rate of spindle burst production above the level seen in the no movement periods, these results suggest some imprecision in the somatotopic organization of barrel cortex at this age, as described previously [
Development of ocular dominance columns in the absence of retinal input.
].
Finally, when expressed as a percentage of the overall spindle burst rate (Figure 4B, dotted lines), the rate of spindle bursts during periods of no movement was 30.6% ± 2.6% during active sleep and 24.8% ± 4.0% during wake. Accordingly, self-generated movements account for approximately 70% of spindle bursts during active sleep and 75% during wake. These values represent our current best estimate of the percentage of spindle bursts in barrel cortex that are related to movement.
Discussion
Before the discovery of retinal waves, researchers debated whether visual system development is “predetermined” (e.g., see [
Retinal waves coordinate patterned activity throughout the developing visual system.
]). Today, this issue is largely settled: there is an overwhelming consensus that the input provided by retinal waves is critical to typical development of the visual system [
Spatiotemporal features of retinal waves instruct the wiring of the visual circuitry.
]. Indeed, rather than debating whether retinal inputs shape the visual system (i.e., whether these inputs are permissive), the field has moved on to asking how patterns of retinal activity lead to normal retinotopy across all levels of the visual system (i.e., how these inputs are instructive; see [
Neuronal activity during development: permissive or instructive?.
]).
In the somatosensory system, however, the debate continues about the importance of peripheral sensory input for cortical development. Whereas some groups posit that self-generated movements, particularly twitches, drive critical early activity in somatosensory cortex [
Early motor activity drives spindle bursts in the developing somatosensory cortex.
], others explicitly state that the sensory feedback from movements is not necessary, instead positing that spontaneous neural activity within the brain is alone sufficient for typical development [
Patchwork-type spontaneous activity in neonatal barrel cortex layer 4 transmitted via thalamocortical projections.
].
It is difficult to imagine how spontaneous neural activity alone could account for the exquisite mapping that develops between body and brain. As is well known, peripheral manipulations of the visual, auditory, and somatosensory systems produce marked changes in cortical representations [
Cortical and thalamic connectivity of the auditory anterior ectosylvian cortex of early-deaf cats: implications for neural mechanisms of crossmodal plasticity.
Early sensory experience influences the development of multisensory thalamocortical and intracortical connections of primary sensory cortices.
]. Nowhere is this clearer than in developing barrel cortex, in which whisker removal and peripherally expressed genetic manipulations reliably alter its structural organization (see [
Development and critical period plasticity of the barrel cortex.
] for review).
Nonetheless, recent reports appear to show that sensory feedback from movements is unrelated to ongoing neural activity within the somatosensory system. There are several factors that can account for our incongruous results. First, we pay close attention to the thermal and hydrational status of our unanesthetized infant subjects and ensure that they are cycling normally between sleep and wake before any data are collected. Second, we use methods that are adequate for detecting movement-related neural activity; in contrast, one recent study used calcium imaging in which data were sampled at only 1 Hz [
Patchwork-type spontaneous activity in neonatal barrel cortex layer 4 transmitted via thalamocortical projections.
], which is inadequate for detecting movement-related activity in barrel cortex (Figure 2D). (In principle, calcium imaging can have sufficient temporal resolution; see [
High-speed in vivo calcium imaging reveals neuronal network activity with near-millisecond precision.
].) By attending to these factors, we have established that the majority of neural activity in barrel cortex is triggered specifically by whisker movements and that sleep is a critical modulator of this process.
By expanding our analysis to define periods when no movement was detected throughout the body, we found that the rate of spindle bursts was only 25% or 31% of that expected during active sleep or wake, respectively (Figure 4B). These values might be interpreted as an estimate of the “spontaneous” or “intrinsic” rate of spindle bursts, that is, spindle bursts that are independent of movement per se. Interestingly, these values are largely consistent with previous studies in which researchers blocked sensory feedback from the whiskers [
Sensory-evoked and spontaneous gamma and spindle bursts in neonatal rat motor cortex.
] and observed 50%–75% decreases in the rate of spindle bursts. The current findings show that, by carefully attending to self-generated movement and behavioral state in intact animals, we can accurately assess the relationships between behavior and early brain activity.
The distinction between sensory input and “intrinsic activity” as contributors to cortical development was extended recently to the late prenatal period in mice (see [
Prenatal activity from thalamic neurons governs the emergence of functional cortical maps in mice.
]). Asserting that this is a period “before sensory input,” the authors concluded that sensory input cannot play a role in the observed cortical map structure. However, the circuitry that enables movement-related sensory feedback is already functional in thalamus by embryonic day 16 [
Prenatal development of spontaneous and evoked activity in the rat (Rattus norvegicus albinus).
]. Thus, it seems likely that sensory feedback drives subcortical neural activity long before the age tested here.
This study has several limitations. First, because we trimmed most of the contralateral whiskers and restrained the forelimbs, we were unable to determine whether movements of the trimmed whiskers and the forelimbs can account for any of the remaining residual activity. Second, we have found that the relationship between neural activity in developing barrel cortex and self-generated movements is probabilistic: Movements increase the likelihood of, but do not determine, a cortical response. In addition to behavioral state and movement amplitude, there are likely other contributing factors that remain to be identified. Finally, we found that cortical activity is greater during active sleep than wake, even during periods when pups were not moving (Figure 4B). This finding suggests a role for state-dependent neuromodulation of corticothalamic excitability, a possibility that remains to be explored.
Altogether, our findings demonstrate that self-generated movements and sleep-wake state critically pattern early somatosensory cortical activity, including spindle bursts. In both visual and somatosensory cortex, the frequency, duration, and topographic specificity of spindle bursts are thought to be crucial in refining cortical topography through at least the first postnatal week [
An excitatory cortical feedback loop gates retinal wave transmission in rodent thalamus.
]. Similarly, by triggering spindle bursts in somatosensory cortex, self-generated movements drive neural activity necessary for somatotopic development. Thus, future research should focus on the specific mechanisms through which patterned somatosensory input enables activity-dependent somatotopic development in this system.
Acknowledgments
This research was supported by grants from the National Institutes of Health ( R37-HD081168 to M.S.B. and F32-NS101858 to J.C.D.). We thank Toby Mordkoff for help with statistical analysis.
Author Contributions
Conceptualization, J.C.D., R.M.G., and M.S.B.; Methodology, J.C.D., R.M.G., G.S., and M.S.B.; Software, J.C.D. and R.M.G.; Validation, J.C.D. and R.M.G.; Formal Analysis, J.C.D. and R.M.G.; Investigation, J.C.D. and R.M.G.; Resources, G.S. and M.S.B.; Data Curation, J.C.D. and R.M.G.; Writing – Original Draft, J.C.D. and R.M.G.; Writing – Review & Editing, J.C.D., R.M.G., G.S., and M.S.B.; Visualization, J.C.D., R.M.G., G.S., and M.S.B.; Supervision, G.S. and M.S.B.; Project Administration, G.S. and M.S.B.; Funding Acquisition, G.S. and M.S.B.
